Skip to content

Pilot ask: advisory ScopeTrail check for agent permission drift #5

Description

@Conalh

Hi. I maintain ScopeTrail, a free local-only GitHub Action that reviews pull requests for risky AI-agent permission/config drift across files like .mcp.json, Claude Code settings, Codex config, and Cursor/VS Code/Windsurf MCP config.This repo looks like a strong real-world pilot candidate because the README documents Claude Code bypassPermissions, Gemini --yolo, and MCP setup through .mcp.json.sample. That is exactly the kind of agent-permission drift ScopeTrail is meant to make visible during PR review.Would you be open to trying ScopeTrail in advisory mode for 3-5 PRs and reporting whether the warnings are useful, noisy, or missing an agent config surface? I am especially interested in whether sample MCP config such as .mcp.json.sample should be surfaced differently from active .mcp.json drift.Recommended trial setup:yaml- uses: Conalh/ScopeTrail@v0.1.11 with: fail-on: noneLinks:- Pilot guide: https://github.com/Conalh/ScopeTrail/blob/main/docs/PILOT.md- Trust docs: https://github.com/Conalh/ScopeTrail/blob/main/docs/TRUST.md- Pilot intake / feedback thread: Conalh/ScopeTrail#18 Structured pilot result form: https://github.com/Conalh/ScopeTrail/issues/new?template=pilot-result.ymlNo PR from me unless you ask for one. If this is not useful for the repo, closing this issue is fine.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions